Deep Learning-Driven Selectivity Enhancement in Synergistic p-Cu ₂ O/n-IGZO Gas Sensor Arrays

We demonstrate a monolithic gas sensor array that integrates p-type Cu2O and n-type a-IGZO films via a UV-assisted precursor patterning method, eliminating the need for etching or development steps. This bidirectional configuration enables p- and n-type sensors to exhibit opposite resistance changes toward the same gas, providing deep learning models with an additional discriminative dimension. The sensor array was evaluated using four representative target gases: ozone (O3), nitrogen dioxide (NO2), hydrogen peroxide (H2O2), and nitrogen monoxide (NO), which include both inorganic oxidizing species and volatile organic compounds. A neural network trained on full resistance–time profiles achieved classification accuracies above 95%, significantly outperforming traditional machine learning algorithms such as support vector machine (76%), random forest (69%), and naïve bayes (50%). Compared to arrays with only a-IGZO sensors (68% accuracy), the inclusion of Cu2O/a-IGZO heterojunctions improved accuracy by over 25%. The system also achieved high-precision gas concentration prediction (R2 > 0.98) and demonstrated excellent humidity tolerance via baseline correction. This scalable, lithography-free strategy offers strong potential for compact and high-selectivity gas sensing systems suitable for portable and real-world environmental monitoring applications.